[论文解读] Human-AI Coevolution

Dino Pedreschi, Luca Pappalardo+15 more|arXiv (Cornell University)|Jun 23, 2023

Complex Network Analysis Techniques被引用 9

一句话总结

本观点型论文倡导社会 AI，勾勒了在复杂系统、网络科学和 AI 的交叉点上研究人类与 AI 在社会技术系统中共同演化的多学科框架，并勾画了开放性问题、方法论和用例。

ABSTRACT

Human-AI coevolution, defined as a process in which humans and AI algorithms continuously influence each other, increasingly characterises our society, but is understudied in artificial intelligence and complexity science literature. Recommender systems and assistants play a prominent role in human-AI coevolution, as they permeate many facets of daily life and influence human choices on online platforms. The interaction between users and AI results in a potentially endless feedback loop, wherein users' choices generate data to train AI models, which, in turn, shape subsequent user preferences. This human-AI feedback loop has peculiar characteristics compared to traditional human-machine interaction and gives rise to complex and often ``unintended'' social outcomes. This paper introduces Coevolution AI as the cornerstone for a new field of study at the intersection between AI and complexity science focused on the theoretical, empirical, and mathematical investigation of the human-AI feedback loop. In doing so, we: (i) outline the pros and cons of existing methodologies and highlight shortcomings and potential ways for capturing feedback loop mechanisms; (ii) propose a reflection at the intersection between complexity science, AI and society; (iii) provide real-world examples for different human-AI ecosystems; and (iv) illustrate challenges to the creation of such a field of study, conceptualising them at increasing levels of abstraction, i.e., technical, epistemological, legal and socio-political.

研究动机与目标

• Motivate a cross-disciplinary study of how AI feedback loops shape collective socio-technical dynamics.
• Define Social AI as the integration of network science, complexity science, and AI to study human-AI interactions at scale.
• Identify open technical and scientific challenges that arise from AI-driven conformism, diversity, and governance in STS.
• Propose methodological directions (interventional, observational, and simulation) to causally assess AI impacts on real users and systems.
• Illustrate the framework with concrete use cases (e.g., navigation systems) and discuss regulatory and ethical considerations.

提出的方法

• Propose a complexity-informed perspective that treats AIs as active shapers of STS via feedback loops.
• Review case studies and phenomena where AI-driven recommendations affect system-level outcomes (e.g., traffic, polarisation, inequality).
• Outline methodological paradigms: model-driven simulations, observational studies, interventional (A/B) trials, and hybrid approaches.
• Discuss data-driven simulations and digital twins to calibrate and test AI impacts on STS.
• Argue for controlled experiments and regulatory-enabled collaborations to assess causal effects of AIs.
• Explore architectural and governance implications for next-generation, diverse, and ethically aligned AI systems.

实验结果

研究问题

• RQ1 How do AI-driven feedback loops alter the structure and dynamics of socio-technical systems compared to pre-AI baselines?
• RQ2 Under what conditions can AIs promote collective benefits (e.g., reduced congestion, less polarisation) rather than conformism or inequalities?
• RQ3 What methodological approaches best establish causal effects of AIs on real users and networks (interventional vs observational)?
• RQ4 What architectures and governance mechanisms support diversity, transparency, and social sustainability in AI-mediated ecosystems?

主要发现

• AI-enabled feedback can amplify collective phenomena (e.g., traffic congestion, polarisation) and may erode diversity if not managed.
• Navigation and recommender systems can yield suboptimal individual outcomes but improved collective metrics when designed with diversity and norms in mind.
• Interventional studies (A/B tests) provide causal evidence of AI impacts, but are resource-intensive and ethically complex.
• Hybrid methods combining observational data with simulations can help bridge gaps where full experimentation is impractical.
• Regulatory frameworks (e.g., EU DSA/DMA) create opportunities for collaborative studies and governance of AI platforms.